Abstract
Objective
We aimed to estimate the prevalence and cumulative incidence of hydroxychloroquine retinopathy (HCQ‐R) and its risk factors among patients receiving long‐term HCQ with rheumatic diseases through a systematic review and meta‐analysis of observational studies that used spectral‐domain optical coherence tomography (SD‐OCT) for screening.
Methods
A systematic search of PubMed, Scopus, Ovid EMBASE, and World Health Organization databases (inception–December 31, 2025) identified observational studies meeting the following criteria: 1) adults with rheumatic diseases on HCQ for at least one year, 2) SD‐OCT used for HCQ‐R screening, 3) reported HCQ‐R prevalence or cumulative incidence or calculable data, 4) reported HCQ‐R risk factors as hazard ratios (HRs) or odds ratios (ORs) with 95% confidence intervals (CIs), and 5) English full text. Study quality was assessed using the Newcastle–Ottawa Scale. Random‐effects meta‐analysis estimated pooled prevalence, cumulative incidence, and risk‐factor associations.
Results
We screened 827 studies from which an overall 19 met inclusion criteria (18 cohort and 1 case‐control). Pooled HCQ‐R prevalence from 2008 to 2023 was 5.1% (95% CI 3.9–6.5). Pooled cumulative incidence was 0.1% (95% CI 0.0–0.5) at 5 years, 2.6% (95% CI 1.6–4.1) at 10 years, and 5.6% (95% CI 3.2–9.6) at 15 years. Risk factors measured as HR (95% CI) included daily dose >5 mg/kg actual body weight (4.32 [2.80–6.65]), chronic kidney disease (CKD) (1.94 [1.27–2.96]), female sex (3.78 [1.90–7.48]), and Asian vs White ethnicity (1.67 [1.07–2.62]).
Conclusion
HCQ‐R risk increases with duration and is dose‐dependent, reaching 5.6% by 15 years. Findings support dose optimization with intensified screening for higher‐risk patients, including those with CKD or those who are female and/or Asian.
INTRODUCTION
Hydroxychloroquine (HCQ) is a generally well‐tolerated antirheumatic medication that is widely used for systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), and other rheumatic diseases, including during pregnancy. 1 , 2 , 3 HCQ improves patient survival 4 , 5 , 6 and reduces disease‐related complications. 7 , 8 , 9 However, it carries a risk of irreversible vision loss due to HCQ retinopathy (HCQ‐R). 1 Earlier screening studies relying on visual field tests and/or clinical examination (eg, bull's‐eye maculopathy) estimated HCQ‐R risk at 0.1% to 0.5% among patients receiving long‐term HCQ. 10 , 11 With the introduction of advanced retinal imaging, particularly macular spectral‐domain optical coherence tomography (SD‐OCT) in 2006, reported HCQ‐R frequency increased, largely because SD‐OCT detects early outer retinal structural changes before typical visual field defects. 12 , 13 , 14
SIGNIFICANCE & INNOVATIONS.
This is the first meta‐analysis to estimate pooled prevalence, cumulative incidence, and risk factors of hydroxychloroquine retinopathy (HCQ‐R) from studies that used spectral‐domain optical coherence tomography for screening.
The weighted pooled prevalence of HCQ‐R was 5% between 2008 and 2023, indicating the overall burden among different screened populations and reflecting both existing and newly detected cases.
Pooled cumulative incidence was 0.1%, 2.6%, and 5.6% at 5, 10, and 15 years, respectively. These results show that risk for HCQ‐R rises over time, which supports the need for continued screening while on HCQ.
Pooled prevalence of 5% mirrored the 15‐year pooled cumulative incidence of 5.6%, which is consistent with the mostly irreversible course of HCQ‐R that cases accumulate instead of recovering over time.
Risk factors for HCQ‐R include patients with an HCQ dose >5 mg/kg/day of actual body weight (ABW), who had a four‐fold higher risk of HCQ‐R compared with those on ≤5 mg/kg/day (ABW). Additional risk factors associated with an increased HCQ‐R risk include chronic kidney disease, female sex, and Asian ethnicity.
In 2011, the American Academy of Ophthalmology (AAO) recommended macular SD‐OCT for chloroquine/HCQ‐R screening and in 2016 updated its guidance to designate SD‐OCT and automated visual field (VF) as the primary screening tests. 15 In the updated AAO recommendation, SD‐OCT is a primary screening test, with wide‐field fundus autofluorescence when available. VF and multifocal electroretinogram (mf‐ERG) are used as confirmatory tests rather than primary screening modalities. 16 Since the adoption of SD‐OCT–based screening recommendations, reported HCQ‐R prevalence has ranged from ~1% to ~15%. 13 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 This variation likely reflects differences in populations, study design, screening practices, and follow‐up. Because HCQ‐R is uncommon and typically late‐onset, robust estimates require large cohorts with long‐term follow‐up; many studies are underpowered to provide precise risks or risk‐factor effect sizes. These evidence gaps may limit patient counseling and optimization of long‐term HCQ monitoring. 28
To address these gaps, we conducted a systematic review and meta‐analysis of observational studies to estimate pooled prevalence and cumulative incidence of HCQ‐R, focusing on studies that used macular SD‐OCT as an objective screening tool. Additionally, we quantified pooled effect sizes for key HCQ‐R risk factors using data from eligible studies.
MATERIALS AND METHODS
This systematic review and the meta‐analysis were performed according to the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines 29 (Supplementary Material 1). We synthesized data from published reports and involved no human participants or identifiable data; therefore, institutional ethics approval was not required for our study. We have registered our systematic review and meta‐analysis on PROSPERO with the registration number CRD42024532847, dated April 19, 2024.
Search strategies
With assistance from a biomedical librarian, ND, and CY searched PubMed, Scopus, EMBASE (Ovid), and the World Health Organization (WHO) Global Index Medicus from inception to December 31, 2025. We used controlled vocabulary (MeSH/Emtree) and free‐text terms for hydroxychloroquine/Plaquenil, retinopathy/retinal toxicity, incidence/prevalence/risk, rheumatic diseases (SLE/lupus, RA), and risk factors. We also hand‐searched reference lists of included full texts and relevant textbooks for additional studies. Search strategies are provided in Supplementary Material 2.
Study selection
We included studies that met the following inclusion criteria: 1) adults on long‐term (≥1 year) HCQ therapy for a rheumatic disease; 2) use of macular SD‐OCT (an objective assessment) with or without another test, such as VF or mf‐ERG (functional test) for HCQ‐R screening, or HCQ‐R identified through guideline‐concordant screenings with clinician‐confirmed diagnoses documented from ophthalmology records during the SD‐OCT era (2006 onward, encompassing early adoption and the 2011 AAO guideline); 3) studies that reported HCQ‐R prevalence during their study period (period prevalence) or data to calculate it; 4) studies that reported cumulative incidence of HCQ‐R since HCQ initiation or data to calculate it; 5) studies on purported risk factors for HCQ‐R, including daily dose, cumulative dose, duration of use, sex, race/ethnicity, chronic kidney disease (CKD), older age or tamoxifen use, which reported their effect sizes as hazard ratio (HR) or odds ratio (OR) with 95% confidence interval (CI) estimates or data to calculate them; and 6) full text in English. Based on our inclusion criteria, cross‐sectional studies, surveys, case‐series, and case‐report study designs were not included. We imposed no restrictions on participant demographics or other study characteristics beyond those prespecified. Two reviewers (ND, CY) independently screened titles/abstracts and full texts, documented exclusion reasons, and resolved disagreements by consensus; unresolved conflicts were adjudicated by a third reviewer (JAAZ).
Data extraction
For studies meeting the eligibility criteria, we extracted the following: 1) study characteristics (authors’ names, year of publication, country of study, cohort type [retrospective vs prospective], study setting [community vs clinic‐based], start and end date of study [study period], follow‐up length [if available], source of information [medical records vs registry vs audit vs health data]; 2) study sample (sample size, number of HCQ‐R detected events [prevalent cases vs incident cases], distribution of age, sex, race/ethnicity, geographic region, and type of rheumatic disease); 3) primary outcome (reported HCQ‐R prevalence, HCQ‐R cumulative incidence at 5, 10, 15, and ≥20 years post‐HCQ initiation), 4) risk factors for HCQ‐R and their effect size with point estimates of HR or OR with (95% CI) or data to calculate them. Information on biologic sex and/or gender were collected when available. Information on race or ethnicity as a purported risk factor for HCQ‐R was categorized per National Institute of Health terminology. 30 ND and CY independently extracted data; discrepancies were resolved by consensus or, if needed, adjudicated by JAAZ.
Risk‐of‐bias (quality) assessment
Two authors (ND, CY) independently applied the Newcastle–Ottawa Scale (NOS), a 10‐point tool for cohort and case‐control studies. Discrepancies were resolved by consensus. Each study was scored on eight items (six scored 0/1; two scored 0/1/2). Scores of 7 to 10 indicated low risk of bias (good), 4 to 6 intermediate risk (fair), and under 4 high risk (poor) quality. For cohort studies, items and scoring were the following: cohort representativeness/setting (community vs clinic), exposure ascertainment/source (medical records/registries vs self‐report), confirmation that HCQ‐R was absent at baseline, selection of participants from the same underlying source population, comparability through confounding control (major and additional covariates), outcome ascertainment during follow‐up (expert confirmation/medical records/International Classification of Diseases codes), adequacy of follow‐up duration (>5 years), and completeness of follow‐up (loss to follow‐up <20%). Full operational definitions and case‐control–specific NOS items are provided in Supplementary Material 3.
Statistical analysis
Period prevalence of HCQ‐R was defined as the proportion of detected HCQ‐R cases during the study period, calculated as the number of existing cases divided by the total number of patients screened within that period. Cumulative incidence of HCQ‐R was the probability that patients free of HCQ‐R at baseline experience HCQ‐R event by time “t” after HCQ initiation, which was defined relative to the baseline at‐risk cohort and estimated as event probability accounting for right censoring and, when applicable, competing risks. 31 , 32
We used a random‐effects model to estimate weighted pooled period prevalence, weighted pooled cumulative incidence, and weighted pooled effect sizes (OR or HR) of the association of risk factors with HCQ‐R. Weighting was based on the study sample size, and pooled estimations were reported as point estimates with 95% CI. With only two studies, we used a fixed‐effects model because between‐study variance is unreliable to estimate for random‐effects. 33
Heterogeneity was assessed using the Q‐test (P ≤ 0.05) and the I2 statistic. 34 To explore potential sources of heterogeneity, we initially conducted subgroup analyses by stratifying on the key methodologic study‐level variables. These included study setting (community vs clinic‐based), cohort type (retrospective vs prospective), study quality (high [≥7] vs low [<7]), publication year (before vs after 2016), follow‐up length (>5 years vs ≤5 years), and geographic region (Europe and North America versus Asia). Although subgroup analyses provided some insight into variation across studies, they were limited in their ability to quantify and adjust for overlapping or confounding effects between variables. Therefore, we proceeded with univariate meta‐regression analysis to further assess the extent to which these methodologic variables contributed to between‐study heterogeneity in logit‐transformed prevalence or cumulative‐incidence estimates. When a methodologic variable demonstrated a significant but partial effect in subgroup analysis, we used multivariable meta‐regression to adjust for the influence of a potential confounding variable and better isolate its independent association. 35
Additionally, we assessed whether study‐level demographics and HCQ exposure variables explained heterogeneity in logit‐transformed prevalence estimates. We tested mean participant age, mean HCQ exposure duration, mean cumulative HCQ dose, and mean daily HCQ dose. Because of the limited number of studies and to avoid overfitting and collinearity, we performed univariate meta‐regression. 35 , 36
To evaluate the robustness of results, leave‐one‐out sensitivity analysis was performed by sequentially removing individual studies and comparing the results with baseline. Our analyses were conducted using Comprehensive Meta‐Analysis Version 4 (Biostat, Englewood, NJ) software.
Publication bias
Publication bias was assessed using funnel plots (SE on the y‐axis and log effect on the x‐axis). In the absence of small‐study effects, studies should be symmetrically distributed around the pooled estimate, with larger studies clustering near the top. Funnel plot asymmetry was further evaluated using Egger's regression test (p value).
Certainty of evidence for outcomes
We assessed the certainty of evidence for each outcome using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. Certainty was rated separately for each of the outcomes. Because all included studies were observational, evidence started at low certainty and was rated down as appropriate for inconsistency (between‐study heterogeneity), indirectness (differences in population, setting, data source), imprecision (width of CI), and publication bias (Egger's test). Final certainty was categorized as low or very low.
RESULTS
Search results
A total of 1,596 records were identified through database searches (Scopus, PubMed, Ovid EMBASE, WHO) and citation tracking. After removing 769 duplicates, 827 records were screened by title and abstract, of which 753 were excluded by consensus for not meeting eligibility criteria. The remaining 74 articles underwent full‐text review. Of these, 55 were excluded for an ineligible study design (ie, cross‐sectional, case‐series, survey) (n = 9), lack of SD‐OCT for screening throughout follow‐up (n = 17), no reported outcome measures or data to calculate (n = 20), ineligible population (nonrheumatic diseases, <1‐year exposure) (n = 6), or non‐English full text (n = 3). Nineteen studies met the inclusion criteria and were included in the risk‐of‐bias assessment and data extraction (Supplementary Figure 1 PRISMA flowchart).
Study characteristics
The main characteristics of the 19 studies included in the systematic review and meta‐analysis are shown in Table 1. The 19 included studies were conducted across diverse geographic regions: 7 from Asia, 17 , 19 , 21 , 22 , 23 , 26 , 37 6 from North America, 13 , 25 , 38 , 39 , 40 , 41 5 from Europe, 18 , 20 , 24 , 27 , 42 and 1 multinational. 43 The majority were retrospective cohorts (14 studies), 13 , 17 , 19 , 21 , 22 , 23 , 25 , 27 , 37 , 41 and almost all used electronic or medical records as data sources. A total of 14 studies were clinic‐based, 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 37 , 41 , 42 whereas 5 were community or population‐based, 13 , 38 , 39 , 40 , 43 reflecting a mix of real‐world practice settings. From the community‐based studies, only one reported prevalence. 13
Table 1.
Characteristics of the 19 studies included in the systematic review and meta‐analysis*
| Author (country) | Design, data/setting, period, FU | Sample (N/n), female (%) | Prev; (Y#) or CumI; (at Y*) | Quality (ROB score) |
|---|---|---|---|---|
| Ahn 2024 (SKR) 17 | Retro cohort, EMR/clinic, 2008–2023, FU ≥5 y | 2,635/92, 94.6 | Prev 3.5%; (15.4 ± 6.8) | Good (7) |
| Alieldin 2023 (UK) 18 | Pros cohort, audit/clinic, 2018–2020, FU <5 y | 566/10, 86 | Prev 1.76%; (9.0 ± 5.0) | Good (7) |
| Almeida‐Brasil 2022 (multi) 43 | Pros cohort (SLICC), RT‐ MR/community, 1999–2019, FU ≥5 y | 1,460/11, 89 | CumI <1%; (10) | Good (8) |
| Dabit 2022 (US) 38 | Retro cohort, health data (REP)/community, 2010–2019, FU ≥5 y | 634/11, 79 | CumI 0% (5); 3.9% (10) | Good (8) |
| Eo 2017 (SKR) 19 | Retro cohort, MR/clinic, 2011–2014, FU <5 y | 310/9, 92.3 | Prev 2.9%; (6.0 ± 3.36) | Fair (6) |
| Gobbett 2021 (UK) 20 | Pros cohort, RT‐MR/clinic, 2017–2019, FU <5 y | 333/2, 85 | Prev 0.6%; (N/A) | Good (7) |
| Jorge 2024 (US) 39 | Retro cohort, health data (KPNC)/community, 2004–2020, FU ≥5 y | 4,677/125, 82.9 | N/A | Good (9) |
| Kao 2022 (TW) 21 | Retro cohort, MR/clinic, 2017–2020, FU <5 y | 92/10, N/A | Prev 10.9%; (11.09 ± 2.44) | Fair (6) |
| Kim 2017 (SKR) 22 | Retro cohort, MR/clinic, 2015–2016, FU <5 y | 123/17, 96 | Prev 13.8%; (10.03 ± 5.08) | Fair (6) |
| Lee 2015 (SKR) 23 | Retro cohort, MR/clinic, 2011–2015, FU ≥5 y | 218/9, 87.6 | Prev 4.12%; (8.37 ± 3.8) | Good (7) |
| Lenfant 2020 (FR) 42 | Retro CC, MR/clinic, a 2007–2010, FU N/A |
|
N/A | Good (8) |
| Marshall 2021 (UK) 24 | Pros cohort, audit/clinic, 2017–2019, FU <5 y | 869/14, 91 | Prev 1.6%; (11.25 ± 5.87) | Good (7) |
| Melles 2014 (US) 13 | Retro cohort (CC analysis), health data (KPNC)/community, 2009–2013, FU ≥5 y | 2,361/177, 83.1 | Prev 7.5%; (12.24 ± 5.04) | Good (9) |
| Melles 2023 (US) 40 | Retrospective, health data (KPNC)/Community, 2004–2020, FU ≥5 y | 3,325/81, 82.7 | CumI 0% (5), 2.5% (10), 8.6% (15) | Good (9) |
| Mukwikwi 2020 (CA) 25 | Retro cohort (CC analysis), registry and MR/clinic, 2016–2019, FU ≥5 y | 326/18, 92 | Prev 5.5%; (17.55 ± 1.62) | Good (7) |
| Petri 2020 (US) 41 | Retro cohort, MR/clinic, 2013–2020, FU ≥5 y | 537/23, 92 | Interval risk b | Good (8) |
| Roy 2021 (IN) 26 | Retro cohort, MR/clinic, 2017–2019, FU N/A | 984/133, 85.8 | Prev 13.5%; (N/A) | Fair (6) |
| Spinelli 2018 (IT) 27 | Retro cohort, MR/clinic, 2008–2016, FU ≥5 y | 504/28, 89.8 | Prev 5.5%; (12.5 ± 6.4) | Good (7) |
| Tangtavorn 2016 (TH) 37 | Retro cohort, MR/clinic, 2004–2014, FU ≥5 y | 61/2, N/A | CumI 3.28%; (10) | Fair (6) |
Y# is mean ± SD of HCQ exposure duration in years; at Y* is years since HCQ initiation. CA, Canada; CC, case‐control; ctrls, controls; CumI, cumulative incidence; EMR, electronic medical records; FR, France; FU, follow‐up; HCQ, hydroxychloroquine; IN, India; IT, Italy; KPNC, Kaiser Permanente Northern California; MR, medical records; Multi, multinational; n, HCQ retinopathy cases; N, total at‐risk; N/A, not available; Prev, prevalence; Pros cohort, prospective cohort; REP, Rochester Epidemiology Project; Retro cohort, retrospective cohort; ROB, risk‐of‐bias; RT‐MR, real‐time MR; SKR, South Korea; SLICC, Systemic Lupus International Collaborating Clinics; TH, Thailand; TW, Taiwan; UK, United Kingdom; US, United States; Y, year.
Multicentral study, controls were taken from PLUS study: randomized, double‐blind, placebo‐controlled, multicentre trial;
Data used to calculate cumulative incidence at to 5, 10, and 15 years post‐HCQ initiation by conducting actuarial survival table method. 32 Interval risk: 1% (≤5), 1.8% (6–10), 3.3% (11–15), 11.5% (16–20), 8% (>20).
Among the 12 studies reporting prevalence, 13 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 female participants comprised 90% of the population. Regarding rheumatic disease types, 59%, 27%, and 14% had SLE, RA, and others, respectively. The mean duration of HCQ use among them was 11.25 ± 3.35 years.
Among studies that reported cumulative incidence in at‐risk patients for HCQ‐R, 37 , 38 , 40 , 41 , 43 86% were female, 60% were diagnosed with SLE, 24% were diagnosed with RA, and the remaining 16% were diagnosed with other rheumatic diseases. The average duration of HCQ exposure was 10 ± 4.33 years.
Weighted pooled estimations
Period prevalence
A total of 12 studies reported the prevalence of HCQ‐R, including 3 prospective cohort studies 18 , 20 , 24 and 9 retrospective cohort studies. 13 , 17 , 19 , 21 , 22 , 23 , 25 , 26 , 27 The pooled sample consisted of 9,321 at‐risk individuals, including 519 prevalent HCQ‐R cases identified across seven countries. The weighted pooled period prevalence from 2008 to 2023 was 5.1% (95% CI 3.9–6.5), based on a random‐effects model (Figure 1). Given substantial heterogeneity (Q = 197.44, P < 0.0001; I2 = 94.4%), we explored potential sources by stratifying studies according to methodologic study‐level variables (Figure 1).
Figure 1.
Forest plot showing period‐prevalence point estimates of HCQ retinopathy for individual studies (open circles; circle size proportional to inverse‐variance weight) and 95% CIs (lines) for individual studies, with the random‐effects pooled prevalence (diamond) per 1,000 population (horizontal bracket denotes the prediction interval). Because heterogeneity was substantial, results were stratified by methodologic variables, including cohort type; prospective vs retrospective; quality score (≥7 vs <7; HQ vs LQ); Pub Y (2016 and later vs before 2016); FU duration (≥5 years vs <5 years); and geographic region (Europe and N America vs Asia). CI, confidence interval; FU, follow‐up; HCQ, hydroxychloroquine; HQ, higher quality; LQ, lower quality; N America, North America; Pub Y, publication year.
Subgroup analysis showed that cohort type (prospective vs retrospective) was the only methodologic variable that explained part of the heterogeneity (P < 0.001). To assess confounding, multivariable meta‐regression showed that retrospective cohorts reported higher prevalence than prospective cohorts after adjusting for follow‐up length (β = 1.72; 95% CI 1.04–2.40; P < 0.001) or publication year (β = 1.69; 95% CI 0.52–2.85; P = 0.0045).
Although higher‐quality studies (≥7) showed a lower pooled prevalence than lower‐quality (<7) studies, the between‐subgroup difference was not statistically significant and did not explain heterogeneity. Likewise, stratification by follow‐up length (<5 vs ≥5 years), publication year (2016 and after vs before 2016), and geographic region (Europe/North America vs Asia) did not materially reduce heterogeneity. Because only one study was community‐based, we did not stratify by setting.
Among demographic/HCQ variables, univariate meta‐regression indicated that only mean daily HCQ dose (β = 0.53; 95% CI 0.14–0.92; z = 2.67; P = 0.007; R2 = 0.57) significantly explained between‐study heterogeneity in prevalence. However, this estimation should be interpreted cautiously because only six studies reported dose metrics, precluding multivariable adjustment.
Cumulative incidence
Four cohort studies reported cumulative incidence of HCQ‐R at 5 years, 38 , 40 and five reported it at 10 years after HCQ initiation 37 , 38 , 40 or data to calculate it, 41 , 43 with three of them additionally providing data at 15 years 40 , 41 , 43 (Figure 2). Weighted pooled cumulative‐incidence estimates were calculated at each time point. The pooled sample comprised 6,017 long‐term HCQ patients from 10 countries, among whom 128 HCQ‐R cases were identified through screening.
Figure 2.
Forest plot of HCQ retinopathy cumulative incidence (per 1,000) at 5, 10, and 15 years after HCQ initiation. Open circles show study‐specific estimates with 95% CIs (horizontal lines); circle size is proportional to inverse‐variance weight. Diamonds indicate random‐effects pooled estimates (5 years: 0.1%; 10 years: 2.6%; 15 years: 5.6%). Horizontal brackets denote prediction intervals. CI, confidence interval; HCQ, hydroxychloroquine. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/acr.80033/abstract.
Weighted pooled cumulative incidence of HCQ‐R was 0.1% (95% CI 0–0.5) at 5 years, 2.6% (95% CI 1.6–4.1) at 10 years, and 5.6% (95% CI 3.2–9.6) at 15 years after HCQ initiation (Figure 2). Heterogeneity among studies within each five‐year interval was nonsignificant (Q‐test = 6.47 for ≤5, 6.85 for 5 to 10, and 3.94 for 10 to 15 years, respectively; P > 0.05). However, I2 values of 53.6%, 41.5%, and 49.2%, respectively, suggested a moderate inconsistency among them. 34
Cumulative incidence of HCQ‐R grouped by daily dosage
Using a random‐effects model, the estimated weighted pooled cumulative incidence of HCQ‐R for daily dose at five years was 0% (95% CI 0–0.3) for doses ≤5 mg/kg actual body weight (ABW) per day compared with 0.1% (95% CI 0–0.7) for doses >5 mg/kg (ABW) per day. At 10 years, the weighted pooled cumulative incidence was 1.1% (95% CI 0.6–2.0) vs 3.7% (95% CI 1.9–7.2), and at 15 years it was 2.7% (95% CI 0.8–8.8) vs 13.8% (95% CI 6.2–28.1), respectively. 40 , 43
Daily dose (HR)
The weighted pooled HR point estimate for HCQ daily dose >5 mg/kg (ABW) vs ≤5 mg/kg (ABW) was 4.32 using the random‐effects model. 38 , 39 , 43
Other covariates (HR)
Pooled HRs for cumulative HCQ dose, CKD, sex, and race/ethnicity are summarized in Table 2. Data on older age at HCQ initiation were inconsistent across the two available studies, 39 , 43 preventing pooled HR estimation (Table 3). Evidence on the risk of HCQ‐R for concurrent use of tamoxifen with HCQ was limited to a single study. 39
Table 2.
Weighted pooled estimation for risk factors of HCQ retinopathy*
| Risk factor | Weighted Pooled HR (95% CI) | Studies (references) |
|---|---|---|
| HCQ daily dose (>5 mg/kg#/day) (reference: ≤5 mg/kg [ABW]/day) | 4.32 (2.80–6.65) | Dabit et al, 38 Jorge et al, 39 Almeida‐Brasil et al 43 |
| Cumulative HCQ dose (per 100‐gr increment) | 1.63 (1.46–1.84) | Dabit et al 38 and Jorge et al 39 |
| CKD (reference: no CKD) | 1.94 (1.27–2.96) | Dabit et al 38 and Jorge et al 39 |
| Female sex (reference: male) | 3.78 (1.90–7.48) | Dabit et al 38 and Jorge et al 39 |
| Race/ethnicity: Asian (reference: White) | 1.67 (1.07–2.62) | Jorge et al 39 and Almeida‐Brasil et al 43 |
ABW, actual body weight; CI, confidence interval; CKD, chronic kidney disease; HCQ, hydroxychloroquine; HR, hazard ratio.
Table 3.
Summary of significant findings on HCQ retinopathy associated risk factors*
| First author and year | Study type | Statistics | Risk factor(s) significantly associated with HCQ‐R risk: point estimates (95% CI) |
|---|---|---|---|
| Almeida‐Brasil et al, 2022 43 | Prospective inception cohort | Unadjusted HR |
|
| Dabit et al, 2022 38 | Retrospective cohort | Unadjusted HR |
|
| Jorge et al, 2024 39 | Retrospective cohort | Adjusted HR |
|
| Kao et al, 2022 21 | Retrospective cohort | Unadjusted OR |
|
| Kim et al, 2017 22 | Retrospective cohort | Adjusted OR |
|
| Lenfant et al, 2020 42 | Case‐control | Adjusted OR |
|
| Melles et al, 2014 13 | Retrospective cohort (case‐control analysis) | Adjusted OR |
|
ABW, actual body weight; CI, confidence interval; CKD, chronic kidney disease; g, gram; HCQ, hydroxychloroquine; HR, hazard ratio; kg, kilogram; mg, milligram; OR, odds ratio; SLE, systemic lupus erythematosus.
Duration of HCQ use (OR)
Two studies provided sufficient data to estimate the association between HCQ duration of use and HCQ‐R prevalence. Using a random‐effects model, the pooled OR per five‐year increase in HCQ use was 2.83 (95% CI 1.26–6.35) (Table 3). 13 , 22
Other covariates (OR)
Three studies provided information to estimate the weighted pooled OR for the risk of HCQ‐R associated with chronic kidney disease (CKD). Using a random‐effects model, the weighted pooled OR of HCQ‐R was 1.87 (95% CI, 0.91–3.85) in those with CKD compared to those without CKD. 13 , 22 , 42 Descriptive statistics regarding risk factors for HCQ‐R from single studies are shown in Table 3. 13 , 21 , 22 , 38 , 39 , 42 , 43
Sensitivity analysis
Leave‐one‐out sensitivity analyses showed that pooled prevalence estimates remained stable when omitting each study in turn (Supplementary Figure 2). Pooled cumulative‐incidence estimates were similarly robust across iterations (Supplementary Figure 3).
Publication bias
Separate funnel plots were generated for prevalence and cumulative incidence (Figure 3A–B). The prevalence funnel plot showed mild asymmetry with greater dispersion among smaller studies, but Egger's test was not significant (P > 0.05). The cumulative‐incidence funnel plot was largely symmetric, and Egger's test was also nonsignificant (P > 0.05), suggesting limited evidence of small‐study effects or publication bias.
Figure 3.
(A) Funnel plot of studies reporting HCQ‐R period prevalence shows mild asymmetry to the left of the vertical line representing the logit of the weighted pooled prevalence; studies toward the bottom have larger standard errors and smaller sample sizes (Egger's regression test P > 0.05). (B) Funnel plot for cumulative incidence is largely symmetrical (Egger's regression test P > 0.05). HCQ‐R, hydroxychloroquine retinopathy.
Risk‐of‐bias (quality) assessment
We assessed risk of bias with the Newcastle–Ottawa Scale; summary scores are presented in Table 1, with item‐level details in Supplementary Material 3.
Certainty of evidence for outcomes
Certainty of evidence (GRADE) was low for the 5‐ and 10‐year cumulative‐incidence estimates and for the effects of dose and CKD. It was very low for period prevalence, 15‐year cumulative incidence, sex, and ethnicity. Details are summarized in Supplementary Table 1.
DISCUSSION
This is the first meta‐analysis to estimate the pooled prevalence, cumulative incidence, and risk factors of HCQ‐R focusing on studies that used SD‐OCT as an objective screening tool to diagnose HCQ‐R. Across studies from 2008 to 2023, pooled prevalence was 5.1%. Pooled cumulative incidence (overall dosing) was 0.1% at 5 years, 2.6% at 10 years, and 5.6% at 15 years after HCQ initiation. Risk was higher with daily dose >5 vs ≤5 mg/kg/day (ABW) (~4‐fold), CKD (~2‐fold), and per 100‐g higher cumulative dose (~1.6‐fold), and was also higher in females (~4‐fold) and Asian versus White patients (~1.7‐fold).
Our pooled prevalence of 5.1% (2008–2023) was similar in magnitude to the pooled 15‐year cumulative incidence of ~5.5%. This is broadly consistent with epidemiologic expectations for a chronic, nonfatal condition with long duration, in which cases accumulate over time and prevalence can approach long‐term cumulative incidence. 31 Prevalence varied widely across studies (ie, 0.6% to 13.8%). 20 , 22 Cohort type appeared to contribute to heterogeneity, with higher prevalence in retrospective studies, a difference that persisted after accounting for follow‐up length and publication year, suggesting design‐related ascertainment differences may partly explain variability. These differences may partly reflect variation in case ascertainment and data completeness across study designs. 44 Definite HCQ‐R requires two abnormal tests (SD‐OCT and visual field), whereas probable HCQ‐R is based on one abnormal test. 15 Retrospective cohorts may misclassify probable as definite because of limited confirmatory testing and may undercount unaffected individuals because of incomplete documentation. Additionally, unmeasured differences in screening frequency and SD‐OCT interpretation thresholds for HCQ‐R diagnosis across cohorts may also contribute to heterogeneity in the observed risk (prevalence or cumulative incidence). 45
Our pooled 10‐year cumulative incidence estimate was similar to that reported in the large community‐based cohort by Melles et al. 40 Differences at 15 years may partly reflect fewer patients remaining at risk in later follow‐up in the single cohort, whereas our estimate is pooled across studies.
Similarly, pooled dose‐stratified cumulative incidence was consistent with large cohorts. 13 , 40 Among patients receiving ≤5 mg/kg/day (ABW), HCQ‐R was rare in the first five years and remained low (1% at 10 years; 3% at 15 years), whereas >5 mg/kg/day was associated with around a four‐fold higher risk (4% at 10 years; 14% at 15 years). These findings highlight the importance of accurate weight‐based dosing, particularly in underweight patients given fixed 200‐mg tablets. Therefore, weekly dose calculations with day‐to‐day tablet adjustments can help maintain recommended limits. In the United States, 100‐mg HCQ tablets became available in 2022, allowing more precise weight‐based dosing. 46
For cumulative incidence, Q‐tests were nonsignificant, but I2 values around 50% indicate moderate inconsistency across time points. 34 Most contributing studies were retrospective cohorts, 37 , 38 , 40 , 41 and only one accounted for the competing risk of death. 38 The remaining study was a prospective multinational Systemic Lupus International Collaborating Clinics (SLICC) inception cohort 43 differing in design, population diversity, and potentially data quality. These clinical and methodologic differences likely contributed to the I2‐based inconsistency despite nonsignificant Q‐tests.
Two community‐based studies 38 , 39 suggested a ~1.6‐fold higher hazard of HCQ‐R per 100‐g increase in cumulative dose. However, high daily dosing over shorter periods may still pose substantial risk despite modest cumulative exposure; HCQ‐R has been reported in graft‐versus‐host disease patients receiving very high mg/kg doses within months to a few years. 47 This information underscores considering daily dose and duration alongside cumulative dose in risk assessment.
Pooled estimates suggest approximately two‐fold higher HCQ‐R risk in patients with CKD, 48 , 49 highlighting the importance of monitoring renal function in those receiving HCQ. For patients with diagnosed CKD, clinicians should consider adjusted weight‐based dosing and initiating screening earlier than five years after HCQ start, with consistent annual screening thereafter. This aligns with the AAO's 2016 recommendations and the recent 2026 revised recommendations for higher‐risk patients. 15
Our pooled estimate suggested that female sex is associated with a higher risk of HCQ‐R (~4‐fold). Because this evidence is limited to two studies with unadjusted 38 versus adjusted 39 analysis, we have to consider that this association may also reflect residual confounding and ascertainment bias rather than intrinsic susceptibility. Biologic sex differences in body size, composition, and fat/lean mass proportions could bias exposure estimates. Fixed tablet dosing can overexpose underweight patients when weight is unmeasured (compromising mg/kg dosing), whereas dosing based on actual body weight in overweight patients may overestimate exposure. 50 Other potential sex‐related contributors include differences in drug metabolism (hormone‐influenced), 51 ocular vascular/structural parameters, 52 , 53 , 54 and differences in medication adherence or screening uptake 55 ; these mechanisms remain plausible but unproven.
Regarding race/ethnicity, studies reported mixed associations for Asian versus White patients, 39 , 43 whereas our pooled estimate suggested higher risk in patients who were Asian and receiving HCQ long‐term. Differential ascertainment may contribute because Asian patients more often have pericentral HCQ‐R that is better detected with wide‐field macular SD‐OCT (12×12 mm) or 24‐2/30‐2 VF testing; thus, guideline‐concordant screening may yield higher observed hazards, whereas underuse may underestimate risk. Biologic differences (eg, retinal melanin content, binding, or structural differences) remain plausible but unproven. 56
Tamoxifen, although not co‐prescribed with HCQ, has been associated with retinopathy (tamoxifen retinopathy), particularly with higher doses and long‐term use. 57 Two large cohorts reported higher HCQ‐R risk with concurrent tamoxifen and HCQ, supporting guideline recommendations to consider tamoxifen as a co‐risk factor. 13 , 39 However, whether tamoxifen or other endocrine therapies (eg, aromatase inhibitors) modify HCQ retinal toxicity remains unclear. 58 Future studies should capture HCQ and endocrine co‐therapies, test effect modification across HCQ exposure metrics, and use guideline‐concordant multimodal outcome ascertainment.
Our study has several limitations. Given substantial heterogeneity in prevalence, we could not conduct multivariable meta‐regression for key study‐level factors because too few studies reported the necessary data. For example, mean daily HCQ dose partly explained heterogeneity in univariate analysis, but dosing data were available in only six studies. As a result, potential confounding among study variables could not be evaluated, and the sources of heterogeneity could not be fully elucidated.
Also, this meta‐analysis evaluated dose using the most commonly reported threshold (>5 vs ≤5 mg/kg/day) and therefore could not examine a graded dose‐response across finer weight‐based categories (eg, <4, 4–5, 5–6, ≥6 mg/kg/day) as suggested in large cohort studies. 13 , 40 We also could not assess HCQ blood levels as a predictor because these data were rarely reported. Given limited reporting of alternative dose cut points and HCQ blood levels despite their importance as potential predictors, future studies are needed to evaluate graded dose‐response relationships and HCQ blood levels in relation to HCQ‐R risk.
Unmeasured heterogeneity, including variation in screening coverage and HCQ adherence, could not be evaluated because of limited reporting. 59 , 60 For example, screening uptake among those receiving long‐term HCQ in England was reported to be only 1.9% to 3.8%, which could underestimate risk. 61 Reported nonadherence to HCQ in SLE ranges widely (3%–75%) depending on assessment method, which may further bias estimates toward under‐ascertainment. 62
In addition, evidence for age at HCQ initiation was limited or inconsistent across studies, with lack of covariate adjustment in most regression analyses. Methodologic heterogeneity further complicates interpretation.
On the other hand, notable strengths of this study include the large, pooled sample size across multiple studies, enhancing the statistical power and generalizability of the findings. Furthermore, the absence of significant publication bias and the stability of results in sensitivity analyses support the robustness of the pooled estimates.
GRADE assessments were largely low to very low, reflecting that the evidence was from observational studies. Certainty was lowest for period prevalence and 15‐year cumulative incidence because of the heterogeneity and wide CI, respectively. Estimates for female sex and race/ethnicity were very low because of sparse data and wide CIs. Associations with 5‐ and 10‐year cumulative incidence, dose, and CKD were rated low certainty.
In summary, this systematic review and meta‐analysis estimated a pooled HCQ‐R prevalence of 5.1%, which is comparable with the 15‐year pooled cumulative incidence (5.6%). Risk increased over time (0.1% at 5 years, 2.6% at 10 years, 5.6% at 15 years) and was strongly dose‐dependent: >5 mg/kg/day (ABW) was associated with almost four‐fold higher hazard and 15‐year incidence of 14% versus 3% at ≤5 mg/kg/day. CKD was a consistent risk factor, with higher risk also observed for female sex and Asian ethnicity among those receiving HCQ long‐term. These findings provide a robust approximation of long‐term risk and underscore the importance of appropriate dosing and adherence to established screening guidelines.
AUTHOR CONTRIBUTIONS
All authors contributed to at least one of the following manuscript preparation roles: conceptualization AND/OR methodology, software, investigation, formal analysis, data curation, visualization, and validation AND drafting or reviewing/editing the final draft. As corresponding author, Dr Avina‐Zubieta confirms that all authors have provided the final approval of the version to be published and takes responsibility for the affirmations regarding article submission (eg, not under consideration by another journal), the integrity of the data presented, and the statements regarding compliance with institutional review board/Declaration of Helsinki requirements.
Supporting information
Disclosure form.
Supplementary Material S1: PRISMA Checklist and Abstract Checklist_merged
Supplementary Material S2: Search Strategy
Supplementary Material S3: Quality score ROB tables
Supplementary Figure 1: PRISMA 2025
Supplementary Figure 2: Sensitivity analysis
Supplementary Figure 3: Sensitivity analysis CumI
Supplementary Table 1: Certainty of Evidence (GRADE) for Summary of Outcomes
ACKNOWLEDGMENTS
We thank Dean Giustini, an experienced biomedical librarian at the University of British Columbia, who assisted authors through our systematic literature search. We thank Dr Mahyar Etminan, PharmD, MSc, an experienced pharmacoepidemiologist, for his constructive comments and guidance throughout the study.
Supported by the Canadian Institutes for Health Research (CIHR) (grant number: PJT‐162133). Dr Daftarian is a recipient of a CIHR Doctoral Research Award (number: 494089). Dr Avina‐Zubieta is the BC Lupus Society Research Scholar and the Walter and Marilyn Booth Research Scholar.
1University of British Columbia, Vancouver, British Columbia, Canada; 2Arthritis Research Canada, Vancouver, British Columbia, Canada; 3Simon Fraser University, Burnaby, British Columbia, Canada.
Additional supplementary information cited in this article can be found online in the Supporting Information section (https://acrjournals.onlinelibrary.wiley.com/doi/10.1002/acr.80033).
Author disclosures are available at https://onlinelibrary.wiley.com/doi/10.1002/acr.80033.
References
- 1. Aviña‐Zubieta JA, Daftarian N, Esdaile JM. Antimalarial medications. In: Wallace DJ, Hahn BH, Askanase A, et al, eds. Dubois’ Lupus Erythematosus and Related Syndromes. 10th ed. Elsevier; 2025:738–749. [Google Scholar]
- 2. Abarientos C, Sperber K, Shapiro DL, et al. Hydroxychloroquine in systemic lupus erythematosus and rheumatoid arthritis and its safety in pregnancy. Expert Opin Drug Saf 2011;10(5):705–714. doi: 10.1517/14740338.2011.566555 [DOI] [PubMed] [Google Scholar]
- 3. Izmirly PM, Costedoat‐Chalumeau N, Pisoni CN, et al. Maternal use of hydroxychloroquine is associated with a reduced risk of recurrent anti‐SSA/Ro‐antibody‐associated cardiac manifestations of neonatal lupus. Circulation 2012;126(1):76–82. doi: 10.1161/CIRCULATIONAHA.111.089268 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Alarcón GS, McGwin G, Bertoli AM, et al; LUMINA Study Group . Effect of hydroxychloroquine on the survival of patients with systemic lupus erythematosus: data from LUMINA, a multiethnic US cohort (LUMINA L). Ann Rheum Dis 2007;66(9):1168–1172. doi: 10.1136/ard.2006.068676 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Canadian Hydroxychloroquine Study Group . A randomized study of the effect of withdrawing hydroxychloroquine sulfate in systemic lupus erythematosus. N Engl J Med 1991;324(3):150–154. doi: 10.1056/NEJM199101173240303 [DOI] [PubMed] [Google Scholar]
- 6. Jorge A, McCormick N, Lu N, et al. Hydroxychloroquine and mortality among patients with systemic lupus erythematosus in the general population. Arthritis Care Res (Hoboken) 2021;73(8):1219–1223. doi: 10.1002/acr.24255 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Petri M. Use of hydroxychloroquine to prevent thrombosis in systemic lupus erythematosus and in antiphospholipid antibody‐positive patients. Curr Rheumatol Rep 2011;13(1):77–80. doi: 10.1007/s11926-010-0141-y [DOI] [PubMed] [Google Scholar]
- 8. Iyer P, Gao Y, Jalal D, et al. Hydroxychloroquine use is associated with reduced mortality risk in older adults with rheumatoid arthritis. Clin Rheumatol 2024;43(1):87–94. doi: 10.1007/s10067-023-06714-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Wasko MCM, Hubert HB, Lingala VB, et al. Hydroxychloroquine and risk of diabetes in patients with rheumatoid arthritis. JAMA 2007;298(2):187–193. doi: 10.1001/jama.298.2.187 [DOI] [PubMed] [Google Scholar]
- 10. Levy GD, Munz SJ, Paschal J, et al. Incidence of hydroxychloroquine retinopathy in 1,207 patients in a large multicenter outpatient practice. Arthritis Rheum 1997;40(8):1482–1486. doi: 10.1002/art.1780400817 [DOI] [PubMed] [Google Scholar]
- 11. Mavrikakis I, Sfikakis PP, Mavrikakis E, et al. The incidence of irreversible retinal toxicity in patients treated with hydroxychloroquine: a reappraisal. Ophthalmology 2003;110(7):1321–1326. doi: 10.1016/S0161-6420(03)00409-3 [DOI] [PubMed] [Google Scholar]
- 12. Marmor MF, Hu J. Effect of disease stage on progression of hydroxychloroquine retinopathy. JAMA Ophthalmol 2014;132(9):1105–1112. doi: 10.1001/jamaophthalmol.2014.1099 [DOI] [PubMed] [Google Scholar]
- 13. Melles RB, Marmor MF. The risk of toxic retinopathy in patients on long‐term hydroxychloroquine therapy. JAMA Ophthalmol 2014;132(12):1453–1460. doi: 10.1001/jamaophthalmol.2014.3459 [DOI] [PubMed] [Google Scholar]
- 14. Browning DJ. The prevalence of hydroxychloroquine retinopathy and toxic dosing, and the role of the ophthalmologist in reducing both. Am J Ophthalmol 2016;166:ix–xi. doi: 10.1016/j.ajo.2016.03.044 [DOI] [PubMed] [Google Scholar]
- 15. Marmor MF, Kellner U, Lai TY, et al; American Academy of Ophthalmology. Recommendations on screening for chloroquine and hydroxychloroquine retinopathy (2016 revision). Ophthalmology 2016;123(6):1386–1394. doi: 10.1016/j.ophtha.2016.01.058 [DOI] [PubMed] [Google Scholar]
- 16. Marmor MF, Ahn SJ, Ehlers JP, et al; American Academy of Ophthalmology. Special AAO report: recommendations on screening for hydroxychloroquine retinopathy (2025 revision). Ophthalmology 2025;S0161‐6420(25)00709–2. [DOI] [PubMed] [Google Scholar]
- 17. Ahn SJ, Kim JH. Screening practices and late diagnosis of hydroxychloroquine retinopathy in Asian patients. Retina 2024;44(4):734–742. [DOI] [PubMed] [Google Scholar]
- 18. Alieldin RA, Boonarpha N, Saedon H. Outcomes of screening for hydroxychloroquine retinopathy at the Manchester Royal Eye Hospital: 2 years’ audit. Eye (Lond) 2023;37(7):1410–1415. doi: 10.1038/s41433-022-02159-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Eo DR, Lee MG, Ham DI, et al. Frequency and clinical characteristics of hydroxychloroquine retinopathy in Korean patients with rheumatologic diseases. J Korean Med Sci 2017;32(3):522–527. doi: 10.3346/jkms.2017.32.3.522 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Gobbett A, Kotagiri A, Bracewell C, et al. Two years’ experience of screening for hydroxychloroquine retinopathy. Eye (Lond) 2021;35(4):1171–1177. doi: 10.1038/s41433-020-1028-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Kao J‐H, Lai T‐T, Lu C‐H, et al. Characteristics and potential risk factors of hydroxychloroquine retinopathy in patients with systemic lupus erythematosus: focusing on Asian population. J Ocul Pharmacol Ther 2022;38(10):728–733. doi: 10.1089/jop.2022.0060 [DOI] [PubMed] [Google Scholar]
- 22. Kim JW, Kim YY, Lee H, et al. Risk of retinal toxicity in longterm users of hydroxychloroquine. J Rheumatol 2017;44(11):1674–1679. doi: 10.3899/jrheum.170158 [DOI] [PubMed] [Google Scholar]
- 23. Lee DH, Melles RB, Joe SG, et al. Pericentral hydroxychloroquine retinopathy in Korean patients. Ophthalmology 2015;122(6):1252–1256. doi: 10.1016/j.ophtha.2015.01.014 [DOI] [PubMed] [Google Scholar]
- 24. Marshall E, Robertson M, Kam S, et al. Prevalence of hydroxychloroquine retinopathy using 2018 Royal College of Ophthalmologists diagnostic criteria. Eye (Lond) 2021;35(1):343–348. doi: 10.1038/s41433-020-1038-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Mukwikwi ER, Pineau CA, Vinet E, et al. Retinal complications in patients with systemic lupus erythematosus treated with antimalarial drugs. J Rheumatol 2020;47(4):553–556. doi: 10.3899/jrheum.181102 [DOI] [PubMed] [Google Scholar]
- 26. Roy AN, Samala V, Kumar YA, et al. Assessing the risk of retinopathy in Indian patients using hydroxychloroquine for rheumatic and musculoskeletal diseases: a retrospective observational study. Indian J Rheumatol 2021;16(1):23–29. doi: 10.4103/injr.injr_185_20 [DOI] [Google Scholar]
- 27. Spinelli FR, Moscarelli E, Ceccarelli F, et al. Treating lupus patients with antimalarials: analysis of safety profile in a single‐center cohort. Lupus 2018;27(10):1616–1623. doi: 10.1177/0961203318781008 [DOI] [PubMed] [Google Scholar]
- 28. Rosenbaum JT, Costenbader KH, Desmarais J, et al. American College of Rheumatology, American Academy of Dermatology, Rheumatologic Dermatology Society, and American Academy of Ophthalmology 2020 joint statement on hydroxychloroquine use with respect to retinal toxicity. Arthritis Rheumatol 2021;73(6):908–911. doi: 10.1002/art.41683 [DOI] [PubMed] [Google Scholar]
- 29. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372(71):n71. doi: 10.1136/bmj.n71 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Lewis C, Cohen PR, Bahl D, et al. Race and ethnic categories: a brief review of global terms and nomenclature. Cureus 2023;15(7):e41253. doi: 10.7759/cureus.41253 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. 3rd ed. Lippincott Williams & Wilkins; 2008. [Google Scholar]
- 32. Barakat A, Mittal A, Ricketts D, et al. Understanding survival analysis: actuarial life tables and the Kaplan‐Meier plot. Br J Hosp Med (Lond) 2019;80(11):642–646. doi: 10.12968/hmed.2019.80.11.642 [DOI] [PubMed] [Google Scholar]
- 33. Borenstein M, Hedges LV, Higgins JPT, et al. When does it make sense to perform a meta‐analysis? In: Borenstein M, Hedges LV, Higgins JPT, Rothstein HR, eds. Introduction to Meta‐Analysis. Wiley; 2009:357–364. doi: 10.1002/9780470743386.ch40 [DOI] [Google Scholar]
- 34. Higgins JP, Thompson SG, Deeks JJ, et al. Measuring inconsistency in meta‐analyses. BMJ 2003;327(7414):557–560. doi: 10.1136/bmj.327.7414.557 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Borenstein M, Hedges LV, Higgins JPT, et al. Subgroup analyses. In: Borenstein M, Hedges LV, Higgins JPT, Rothstein HR, eds. Introduction to Meta‐Analysis. Wiley; 2009:149–186. doi: 10.1002/9780470743386.ch16 [DOI] [Google Scholar]
- 36. Borenstein M, Hedges LV, Higgins JPT, et al. Meta‐regression. In: Borenstein M, Hedges LV, Higgins JPT, Rothstein HR, eds. Introduction to Meta‐Analysis. Wiley; 2009:187–203. doi: 10.1002/9780470743386.ch17 [DOI] [Google Scholar]
- 37. Tangtavorn N, Yospaiboon Y, Ratanapakorn T, et al. Incidence of and risk factors for chloroquine and hydroxychloroquine retinopathy in Thai rheumatologic patients. Clin Ophthalmol 2016;10:2179–2185. doi: 10.2147/OPTH.S119872 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Dabit JY, Hocaoglu M, Moder KG, et al. Risk of hydroxychloroquine retinopathy in the community. Rheumatology (Oxford) 2022;61(8):3172–3179. doi: 10.1093/rheumatology/keab844 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Jorge AM, Melles RB, Marmor MF, et al. Risk factors for hydroxychloroquine retinopathy and its subtypes. JAMA Netw Open 2024;7(5):e2410677–e2410677. doi: 10.1001/jamanetworkopen.2024.10677 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Melles RB, Jorge AM, Marmor MF, et al. Hydroxychloroquine dose and risk for incident retinopathy: a cohort study. Ann Intern Med 2023;176(2):166–173. doi: 10.7326/M22-2453 [DOI] [PubMed] [Google Scholar]
- 41. Petri M, Elkhalifa M, Li J, et al. Hydroxychloroquine blood levels predict hydroxychloroquine retinopathy. Arthritis Rheumatol 2020;72(3):448–453. doi: 10.1002/art.41121 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Lenfant T, Salah S, Leroux G, et al; PLUS Group . Risk factors for hydroxychloroquine retinopathy in systemic lupus erythematosus: a case‐control study with hydroxychloroquine blood‐level analysis. Rheumatology (Oxford) 2020;59(12):3807–3816. doi: 10.1093/rheumatology/keaa157 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Almeida‐Brasil CC, Hanly JG, Urowitz M, et al. Retinal toxicity in a multinational inception cohort of patients with systemic lupus on hydroxychloroquine. Lupus Sci Med 2022;9(1):e000789. doi: 10.1136/lupus-2022-000789 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Jurek AM, Greenland S, Maldonado G, et al. Proper interpretation of non‐differential misclassification effects: expectations vs observations. Int J Epidemiol 2005;34(3):680–687. doi: 10.1093/ije/dyi060 [DOI] [PubMed] [Google Scholar]
- 45. Kim J, Kim KE, Kim JH, et al. Practice patterns of screening for hydroxychloroquine retinopathy in South Korea. JAMA Netw Open 2023;6(5):e2314816. doi: 10.1001/jamanetworkopen.2023.14816 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. National Library of Medicine (US) . Hydroxychloroquine sulfate tablet. DailyMed. 2025. Accessed June 1, 2026. https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=84b00366-96ef-41e1-bac6-5d24acdb9e1d [Google Scholar]
- 47. Navajas EV, Krema H, Hammoudi DS, et al. Retinal toxicity of high‐dose hydroxychloroquine in patients with chronic graft‐versus‐host disease. Can J Ophthalmol 2015;50(6):442–450. doi: 10.1016/j.jcjo.2015.08.003 [DOI] [PubMed] [Google Scholar]
- 48. Furst DE. Pharmacokinetics of hydroxychloroquine and chloroquine during treatment of rheumatic diseases. Lupus 1996;5(suppl 1):S11–S15. doi: 10.1177/0961203396005001041 [DOI] [PubMed] [Google Scholar]
- 49. Tett SE, Cutler DJ, Day RO, et al. Bioavailability of hydroxychloroquine tablets in healthy volunteers. Br J Clin Pharmacol 1989;27(6):771–779. doi: 10.1111/j.1365-2125.1989.tb03439.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50. Browning DJ, Lee C. Somatotype, the risk of hydroxychloroquine retinopathy, and safe daily dosing guidelines. Clin Ophthalmol 2018;12:811–818. doi: 10.2147/OPTH.S163451 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 51. Yang L, Li Y, Hong H, et al. Sex differences in the expression of drug‐metabolizing and transporter genes in human liver. J Drug Metab Toxicol 2012;3(3):1000119. doi: 10.4172/2157-7609.1000119 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52. Iwase T, Yamamoto K, Yanagida K, et al. Investigation of causes of sex‐related differences in ocular blood flow in healthy eyes determined by laser speckle flowgraphy. Sci Rep 2017;7(1):13878. doi: 10.1038/s41598-017-14118-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53. Li XQ, Larsen M, Munch IC. Subfoveal choroidal thickness in relation to sex and axial length in 93 Danish university students. Invest Ophthalmol Vis Sci 2011;52(11):8438–8441. doi: 10.1167/iovs.11-8108 [DOI] [PubMed] [Google Scholar]
- 54. von Hanno T, Lade AC, Mathiesen EB, et al. Macular thickness in healthy eyes of adults (N = 4,508) and relation to sex, age and refraction: the Tromsø Eye Study (2007‐2008). Acta Ophthalmol 2017;95(3):262–269. doi: 10.1111/aos.13337 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 55. Vaidya V, Partha G, Karmakar M. Gender differences in utilization of preventive care services in the United States. J Womens Health (Larchmt) 2012;21(2):140–145. doi: 10.1089/jwh.2011.2876 [DOI] [PubMed] [Google Scholar]
- 56. Yusuf IH, Charbel Issa P, Ahn SJ. Hydroxychloroquine‐induced retinal toxicity. Front Pharmacol 2023;14:1196783. doi: 10.3389/fphar.2023.1196783 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 57. Tenney S, Oboh‐Weilke A, Wagner D, et al. Tamoxifen retinopathy: a comprehensive review. Surv Ophthalmol 2024;69(1):42–50. doi: 10.1016/j.survophthal.2023.07.003 [DOI] [PubMed] [Google Scholar]
- 58. Sit M, Mason RH, Kaplan AJ. Hydroxychloroquine ocular toxicity: a comprehensive review. J Rheumatol 2026;jrheum.2025‐0127. [DOI] [PubMed] [Google Scholar]
- 59. Deeks JJ, Higgins JPT, Altman DG, et al. Analysing data and undertaking meta‐analyses. In: JPT Higgins, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA, eds. Cochrane Handbook for Systematic Reviews of Interventions. Version 6.5. Cochrane; 2024. Updated November 2024. [Google Scholar]
- 60. Käyser SC, Dekkers T, Groenewoud HJ, et al. Study heterogeneity and estimation of prevalence of primary aldosteronism: a systematic review and meta‐regression analysis. J Clin Endocrinol Metab 2016;101(7):2826–2835. doi: 10.1210/jc.2016-1472 [DOI] [PubMed] [Google Scholar]
- 61. Yusuf IH, Han RC, Downes SM, et al. A nationwide survey of hydroxychloroquine retinopathy presenting to the hospital eye service in the United Kingdom. Eye (Lond) 2023;37(10):2082–2088. doi: 10.1038/s41433-022-02291-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 62. Costedoat‐Chalumeau N, Amoura Z, Hulot JS, et al. Very low blood hydroxychloroquine concentration as an objective marker of poor adherence to treatment of systemic lupus erythematosus. Ann Rheum Dis 2007;66(6):821–824. doi: 10.1136/ard.2006.067835 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Disclosure form.
Supplementary Material S1: PRISMA Checklist and Abstract Checklist_merged
Supplementary Material S2: Search Strategy
Supplementary Material S3: Quality score ROB tables
Supplementary Figure 1: PRISMA 2025
Supplementary Figure 2: Sensitivity analysis
Supplementary Figure 3: Sensitivity analysis CumI
Supplementary Table 1: Certainty of Evidence (GRADE) for Summary of Outcomes